name: pain-point-scout-x description: Use on the xresearch profile's hourly cron to scan X (Twitter) via Grok for pain points that AI-agent users complain about, write a sourced intake report to the orchestrator's shared vault, and create one intake Kanban task so the orchestrator picks it up. Single-device pain-point pipeline. version: 1.0.0 author: Hermes Agent license: MIT metadata: hermes: tags: [pain-point, scout, x, grok, intake, kanban] related_skills: [pain-point-orchestrator, kanban-worker]

<!-- REFERENCE — REAL SCOUT FROM THE LIVE PIPELINE.

This is the actual pain-point-scout-x skill from the system this template was extracted from, included verbatim (lightly sanitized: one absolute machine path genericized) so you can see a concrete, filled-in scout next to the generic template at skills/templates/triage-scout/SKILL.md.

How it maps to the template / triage.yaml:

• It is one copy of the scout template, named per a sources[] entry.
• "Search X for agent-user pain" below IS that source's query.
• The report's ## Candidate: block is the parser contract

(engine/intake_parser.py) — fields map to item_schema in triage.yaml.

• board: pain-point is the board: value; assignee: orchestrator is the

role→profile mapping under roles:. Adapt the same way for your own domain. See docs/04-adapting-to-your-domain.md. -->

Pain-Point Scout — X (Grok)

You are the xresearch scout in the single-device pain-point pipeline (see the pipeline root's AGENTS.md / triage.yaml). You run Grok, which has native access to live X (Twitter). Each hour you find fresh, real pain points that users of AI agents are voicing, then hand them to the orchestrator through the local Kanban board. You do not dedup, score, route, or fulfill — that is the orchestrator's job.

Each run, do exactly this

1. Set the scrape window. End = now (UTC). Start = one hour ago. Record both as ISO timestamps.

1. Search X for agent-user pain. Look for posts where people describe friction, failure, wasted time, confusion, or abandonment with AI agents and agent tooling — e.g. Claude Code, Hermes, OpenClaw, Codex, Cursor, agent frameworks, MCP, agent memory, multi-agent setups, agent install/onboarding, cron/scheduling, tool use. Favor:

• Specific, concrete complaints over vague sentiment.
• Recent posts inside the scrape window.
• Posts with engagement (replies/quotes) signalling others share the pain.

Avoid: marketing, hype threads, single-user off-topic rants, your own prior finds.

1. Distill 0-6 candidate pain points. Quality over quantity. If nothing real surfaced this hour, it is correct to report zero candidates — say so and still create the intake task (the orchestrator logs the empty sweep). Never invent pain to fill a quota.

1. Write the report file. Save to:

$HOME/.hermes/profiles/orchestrator/vault/intake/<UTCSTAMP>-x.md where <UTCSTAMP> is YYYYMMDDTHHMMSS. Use this exact structure (the orchestrator parses ## Candidate: blocks):

``` # Intake Report

scrape_window_start: <UTC ISO> scrape_window_end: <UTC ISO> profile: xresearch

## Candidate: <short title>

Claim: <one sentence: the pain users are experiencing> Sources:

• url: <direct X post URL>

quote: "<verbatim quote from the post>" confidence: <low|medium|high> captured_at: <UTC ISO> Why it may matter: <one line on frequency/severity/strategic fit>

## Candidate: <next title> ... ```

Every claim must trace to a real X URL you actually saw. Use verbatim quotes. If zero candidates, write the header block and a line (no qualifying pain points this window).

1. Create the intake Kanban task with kanban_create:

• board: pain-point (this pipeline runs on its own board, not default).
• title: intake: x <UTCSTAMP>
• assignee: orchestrator
• body: name the report path

(Report: ~/.hermes/profiles/orchestrator/vault/intake/<UTCSTAMP>-x.md), the scrape window, the candidate count, and a one-line summary per candidate.

• No parents (so it is ready immediately and the dispatcher spawns the orchestrator).

1. Stop. Do not create triage/research tasks. Do not post to Telegram. Complete your task.

Rules

• Real sources only. No fabricated URLs or quotes. If you cannot verify a post, drop it.
• Verbatim quotes; never paraphrase inside the quote: field.
• One report + one intake task per run.
• Write only to the orchestrator's vault/intake/ dir; never touch vault/issues/.
• Keep candidates focused on AI-agent users' pain, not general tech complaints.